I have a set of items with their attributes. I want to use this data to train a an open-source model that uses users-items interactions as positive preferences data.
Would it be a good idea to consider I have only one global user to train the model, even if I think it will not consider no/negative preferences? Or to randomly split the items and allocate them to virtual users (which would create incoherence in the preferences)?
Is there any SOTA I can use to simulate users' preferences based on the items' attributes?
Any pointer is very welcomed!